DE-2LS boosts constrained optimization with late-stage refinement
New DE-2LS algorithm slashes optimization costs by 5.58% while improving accuracy
Researchers Dikshit Chauhan and Anupam Trivedi have developed DE-2LS, a new constrained numerical optimization algorithm that enhances the RDEx framework with a lightweight late-stage local search component.
The method maintains all original RDEx features (mutation, crossover, success-history adaptation, archive mechanism) while adding a coordinate-pattern local search that activates only in the final optimization stages using a small evaluation budget. This approach differs from traditional methods by avoiding aggressive early-stage refinement in favor of controlled polishing when the algorithm nears convergence. In comparative tests, DE-2LS achieved a 5.58% U-score improvement over its base RDEx implementation and secured the highest overall score of 80,968 among competing algorithms (RDEx, CL-SRDE, UDE-III). The source code has been released on arXiv for public use.
- DE-2LS combines differential evolution with late-stage local search for constrained optimization
- Achieved 5.58% U-score improvement over RDEx and highest total score of 80,968 in benchmark tests
- Lightweight polishing component activates only in final stages using small evaluation budget
Why It Matters
DE-2LS delivers more accurate optimization results at lower computational cost for engineering and AI applications